Load balancing and unbalancing for power and performance in cluster-based systems

In this paper we address power conservation for clusters of workstations or PCs, such as those that support a large number of research/teaching organizations and most Internet companies. Our approach is to develop systems that dynamically turn cluster nodes on – to be able to handle the load imposed on the system efficiently – and off – to save power underlighter load. The key componentof our systems is an algorithm that makes load balancing and unbalancing decisions by considering both the total load imposed on the cluster and the power and performance implications of turning nodes off. The algorithm is implemented in three different ways: (1) at the application level for a cluster-based, locality-conscious network server; (2) at the operating system level for an operating system for clustered cycle servers; and (3) by application/operating system negotiation, again using the cluster-oriented operating system and a negotiation API. Our experimental results are very favorable, showing that our systems conserve both power and energy in comparison to traditional systems, which do not adapt their configurations.Technical report DCS-TR-44